topic Re: Performance issue with alfresco 5.0d and solr4 in Alfresco Forum

Performance issue with alfresco 5.0d and solr4

ecarbenay — Fri, 13 Jan 2017 16:02:17 GMT

Hello,we're having a strange issue with alfresco 5.0 d and SOLR4.The server is running with this alfresco version since one year, without any problem.Alfresco has been installed manually, by deploying the different war after installing tomcat and mysql.Alfresco is running behind a reverse proxy (an

Re: Performance issue with alfresco 5.0d and solr4

afaust — Fri, 13 Jan 2017 16:16:54 GMT

Looks like SOLR is doing full text indexing. If there is a misconfiguration in the SOLR tier that results in SOLR not finding its old (moved) index files, it will simply start and create a new index from scratch.

The SOLR configuration files seem to be fine - I thought I saw an issue for a moment but turned out I just misread something due to the bad default text reader for zipped content.

Re: Performance issue with alfresco 5.0d and solr4

ecarbenay — Tue, 17 Jan 2017 17:24:00 GMT

Thank you for your answer Axel.

I understand that to your opinion the configuration used is the good one regarding SOLR.

Despit that the load on the server continues to be heavy, and we get no explanation about that issue.

Do you have some recommendation to search the reason about that heavy load ? We think this heavy load is caused by SOLR because of the number of call to textContent (/service/api/solr/textContent) which represents 86 % of the activities on the server, and the X509 Authentication failure we get when running textContent manually. We don't find any way to check why we get such X509 when running it when it seems Alfresco is able to run it because documents are content indexed. Do you have some ideas about that ?

Re: Performance issue with alfresco 5.0d and solr4

afaust — Tue, 17 Jan 2017 20:10:01 GMT

I already mentioned one possible cause: If the /data/solr4 directory did not contain the old index data after the move, then SOLR started a new indexing process causing as many calls to /textContent as there are documents. How many documents do you have stored in that system? How many of those documents may need to be transformed into text first (i.e. Office documents / images to be OCR'ed)?

In the screenshot of top there really isn't that much CPU load on the user side - most of the load is on IO wait times which means that some storage device / filesystem level access is causing the load.

It /data/solr4 mapped to any kind of special drive / storage area? How is the general performance of that device, i.e. what is its latency / throughput rating?

Re: Performance issue with alfresco 5.0d and solr4

tdaget — Thu, 04 May 2017 08:37:37 GMT

Hi all,
I'm working with Emmanuel CARBENAY about this problem. It's closed right now as it doesn't occur anymore.

We are not absolutely confident to have the right solution (maybe there is somethink else to focus to) but the slow down symptom doesn't appear at all since end of january ...

The solution lies in the MySql configuration. Too many database connections was allowed consuming a lot of memory. As some free memory was missing, tomcat try to free it too often, making system starting to swap and the overall platform go to slow down. It was occuring during ~15min at a random point in the day (maybe while final users try to work a little more ?).
Decrease 'max_connections' parameter was the first good thing to do. Alfresco 'db.pool.max' parameter has to be set according ...

We found this solution based on global memory usage analyse and detailed unix process memory consumtion analyse ... it was not funny and not trivial ...

Thanks for your answer
Thomas